A I -D A –I I R Tapps.aima.in/ejournal_new/articlesPDF/3-Pooja-Sharma.pdfMany other studies have also defined AI in a similar manner comprising technologies like - speech recognition,

1AIMA Journal of Management & Research, February 2020, Volume 14 Issue 1/4, ISSN 0974 – 497Copy right© 2020 AJMR-AIMA

Article No. 3

ARTIFICIAL INTELLIGENCE-DRIVENPROCESS AUTOMATION – IS IT REAL THIS

TIME

Pooja SharmaSenior Consultant, Tata Consultancy Services Ltd. & Ph.D. Scholar (IGNOU)

Dr. Nayantara PadhiAssociate Professor, School of Management Studies, IGNOU, New Delhi

Abstract: The disruptive potential of AI is being compared with that of electricity or steamengine. While automation of tasks and processes in banking is not a new concept, the pervasive natureof automation using intelligent technologies is on a different scale. This paper is based on extensiveliterature review and qualitative data gathered by the researcher from experts and practitioners in thefields of automation and banking. The article begins with building a logical yet easy to comprehendtechnology framework for process automation in banking. It further demonstrates the convergenceeffect of the framework using a sample business process in the retail lending space. The articleconcludes with establishing the real-nature of the technological impact on banking automation andhopes to assist the bankers in understanding this complex technological space.

Keywords: Intelligent automation, Convergence, Process, Banking

Introduction

“ARTIFICIAL INTELLIGENCE [Al] and other developments in computer scienceare giving birth to a dramatically different class of machines that can perform tasksrequiring reasoning, judgment, and perception that previously could be done only byhumans. Will these machines reduce the need for human toil and thus causeunemployment?” (Nilsson, 1984)

In their 2004 book titled ‘The New Division of Labor: How Computers Are Creatingthe Next Job Market,’ noted MIT and Harvard professors Frank Levy and Richard JMurane had pointed out the near impossibility of replicating human cognitiveprocesses such as driving in traffic (Brynjolfsson & McAfee, 2011). And yet inFebruary 2018, the state of California, USA, has passed a law where for the first time,companies will be able to operate autonomous vehicles in public places without asafety driver behind the wheel (Hafner, 20). In 2011, the author had presented a paperrecommending the of futuristic technologies such as AI and robotics for low-endroutine tasks as well as supervisory work (Jayanthi & Pooja, 2011).

Evidence suggests that artificial intelligence technologies are coming out of hypemode of more than three decades due to other enabling ecosystem developments. It isalso suffering from overuse and misplaced use in many cases due to its highlycomplex nature. This article seeks to deconstruct the technical mystery surroundingAI. It also seeks to evaluate the relative maturity levels of the basket of AI-relatedtechnologies, which are contributing to a technologically conducive ecosystem. AI is


termed as a general-purpose technology; however, we limit our discussion here toonly to process automation in banking. In the process, the authors create a uniqueconceptual framework focussed on intelligent process automation in banking as a usecase. The authors provide evidence of the usage of the framework with one commonuse-case from banking, which is unfolding in the current times. The objective is tospawn more detailed process reviews and discussions in the banking community toproactively address these impending changes.

DECONSTRUCTING ARTIFICIAL INTELLIGENCE

AI, digital, automation though different terms, are often used interchangeably andrepresent numerous technologies demonstrating certain key characteristics. Generally,AI comprises technologies such as deep learning, neural networks, natural languageprocessing (NLP), machine learning, robotics & cognitive computing. Many times, AIis treated synonymous with cognitive systems or autonomous systems as well sincethey imbibe human-like traits of reasoning, problem-solving, learning, judgment, anddecision-making. Many other studies have also defined AI in a similar mannercomprising technologies like - speech recognition, NLP, semantic technology,biometrics, machine & deep learning, swarm intelligence and chatbot/voice bots(Stancombe et al., 2017), (Dawar & Lacy, 2017), (McKinsey, 2017).

An illustrative list of some of the popular technologies comprising or leading up to AIand relevant in the context of process automation is presented in Table 1.Technology ExampleArtificial Intelligence Machine Learning, Supervised Learning,

Transfer Learning, Cognitive Computing,etc.

Neural Networks Artificial Neural Network, DeepLearning, etc.

Robotics Soft Robotics, Humanoid robotics, etc.Automation Product Categories Unmanned aerial vehicles, Chatbots,

Robotic Process Automation, etc.Table 1: AI Technologies

Source: (McKinsey, 2017)

Another useful way of defining Artificial intelligence in the context of automation canbe derived based on task characteristics such as one taken from a recent TCS report.The report defines AI as those set of technologies that can perform the following fourtasks:

“Sense – Being able to recognize images, sound, voice, video, andother ’unstructured’ data (as well as structured data that has appeared incomputer databases for years)Think – Deciding what such digital data means, and doing so at lightning speedbased on algorithmsAct – Determining what to do about insights after arriving at themLearn – Being able to continuously and automatically refine the knowledge andalgorithmic models of an AI system based on its interactions with digital data;increasingly, such learning is referred to as ‘machine learning.’” (TCS, 2017)


The discussions and arguments in this article leverage these concepts and definitionsto build the progress, placement, and contribution of the various components of anAI-driven process automation framework -‘‘The Great Convergence Framework(GC Framework”.While it is too early to see any mass-scale implementations of theframework and/or its variants, the authors take reference of one use-case frombanking to demonstrate the deployment of the framework in real-life situations as aprecursor to the future state of process automation in banks.

METHODOLOGY

Due to the forward-looking nature of the topic, and the lack of specific empiricalstudies in this context, we follow a rather non-traditional approach in this article. Wedo not report new data, demonstrate the existence of a new variable, or testhypotheses. Instead, the principal contribution of this article is to bring out linkagesamong seemingly disparate technological developments to create a cohesive, logicalframework – GC Framework.

The GC framework is embedded in an extensive literature review. Over 50 articles,research studies, journal papers from traditional and non-traditional sources werescoured to provide the technology base for the framework. The findings were adaptedto process automation in banking and validated for the Indian banking context viaqualitative content analysis of face-to-face in-depth interviews from 28 experts bothfrom the supply-side and demand side of the automation equation. The supply-siderespondents comprise implementers, consultants, system integrators in the area ofprocess automation, and the demand side respondents comprise senior bankersinvolved in either implementing the technology or using the same in businessprocesses. A detailed profile of the participants in the study (n=28) are providedbelow:

S.No. Code PrimaryCategory

Role Organisation Profile

1 IN01 Bank Banking Consultant Large public sector bank2 IN02 Bank Ex-Generall Manager IT Large public sector bank3 IN03 Bank Deputy General Manager Large public sector bank4 IN04 Bank Deputy General Manager Large public sector bank5 IN05 Bank Deputy General Manager Large public sector bank6 IN06 Bank Vice President New private sector bank7 IN07 Bank Vice President New private sector bank8 IN08 Bank General manager Reserve Bank of India9 IN09 Bank Zonal Head New private sector bank10 IN10 Bank CEO Business Correspondent11 IN11 Bank Deputy General Manager New private sector bank12 IN12 Bank Chief Operating Officer New private sector bank13 IN13 Bank Assistant General Manager Large public sector bank14 IN14 Bank Assistant General Manager Large public sector bank15 IN15 Bank Assistant General Manager Large public sector bank16 IN16 Technology Chief Technology Officer,

Banking & FinancialLarge IT/ITES company


Services

17 IN17 Technology Global Head - DigitalTransformation

Large IT/ITES company

18 IN18 Technology Assistant Vice President Large IT/ITES company19 IN19 Technology Founder & CEO Fintech20 IN20 Technology Ex-Head BPO Large IT/ITES company21 IN21 Technology IT Consultant & Member Task Force on Artificial

Intelligence for India’sEconomicTransformation,Ministry of Commerce& Industry

22 IN22 Technology CEO Fintech23 IN23 Technology Head Corporate Strategy Large IT/ITES company24 IN24 Technology IT Consultant Large IT/ITES company25 IN25 Technology Chief Technology Officer,

Banking & FinancialServices

Large IT/ITES company

26 IN26 Technology Assistant Vice President Large IT/ITES company27 IN27 Technology Vice President Large IT/ITES company28 IN28 Technology CEO Fintech

The outcome is an internally consistent analysis that, while firmly grounded inpublished work, creates a new business-friendlyly intelligent process automationframework.

CONCEPTUAL MODEL – TECHNOLOGY SELECTION, GROUPING, ANDPLACEMENT

The critical input to various technologies in the ecosystem comes from the literaturereview. The development of individual technologies is well established in theliterature. The proposed framework, though embedded in technology elements, needsto bring out the business impact of one or more of these technology elements workingtogether. For this purpose, we also need to assess the selection of technologiescomprising the framework. The following six specific frameworks were studied andcompared for relevance.

1. TOE Framework – Technology – Organisation – Environment

2. Diffusion of Innovations

3. Technology Readiness Level (TRLs) originally conceptualised by NASA

4. Gartner Hype Cycle and technology maturity levels

5. Reinforcement Loops

6. Motivation Theory


(Gartner Inc, 2018) , (Gartner Inc., 2017) , (Gartner Inc., 2017) , (Gangwar & Date,2015) , (HOTI, 2015) , (Botha & Atkins, 2005) , (Olechowski, Eppinger, & Joglekar,2015)

Each of these frameworks is industry standard used in varying scenarios, whether it betechnology adoption in enterprises (covered by TOE and Diffusion of Innovations), orassessment of maturity level of specific technologies (TRLs and Gartner hype cycles),etc. It has to be noted here that no one theoretical framework for innovationimplementation or technology assessment covers complex technologies and systemssuch as the ones selected for our GC Framework. However, we do take inspirationfrom the Gartner Hype Cycle and technology maturity levels to validate the selectionof the technologies included in the GC framework.

We studied the Gartner Hype Cycles for some of the technologies selected for the GCframework (Gartner Inc., 2017), (Gartner Inc, 2018), (Gartner Inc., 2017). To oursatisfaction, we found almost all the technologies placed in the enablement layer areout of the ‘embryonic’ phase and either in the ‘emerging,’ ‘adolescent’ or ‘earlymainstreaming’ phase according to Gartner (detailed explanation of the maturitylevels is given in the Annexure1). This means that the technologies selected for theframework have entered industries and businesses as technology solutions. Thiscreates a sound foundation for the proposed framework.

The placement and grouping of technologies for bank process automation was doneusing the primary data collected from the interviews in response to the followingquestions:

Question (1): “We have seen in the past automation in banking services have eitherpushed tasks towards a self-service model, e.g., delivery channels or has codifiedsome parts of rule-driven processes, e.g., credit scoring. In your view, what isdifferent this time around with industry talking about ‘intelligent automation’ from atask and process perspective in banking?”

Question (2): “I will list some categories of popular automation technologies. Pleaserate their importance in its potential impact on Indian banking in the next ten years?”

Question (3): “Please provide any other technology which is not stated above but isvery important in banking automation?”

The findings and results have been discussed below.

FINDINGS & RESULTS - ‘THE GREAT CONVERGENCE’ FRAMEWORK

In 2017, Accenture and Digital Transformation Initiative (DTI) of the WorldEconomic Forum (WEF) published a report evaluating the impact of technology-based innovation on jobs across ten industries. The report charts out the ‘cumulativetechnological capability’ available today due to the simultaneous peaking of varioustechnologies. The time graph plot in the report clearly shows the demise of oldertechnologies like mainframes & client-server, and a convergence high towards 2020,for digital technologies including AI, IoT, smart machines, big data, etc. (Dawar &


Lacy, 2017). A similar sentiment is echoed in a paper by the IBA Global EmploymentInstitute,

“the term Industry 4.0 comprises the technical integration of cyber-physicalsystems (CPS) into production and logistics and the use of the ‘internet ofthings’ (connection between everyday objects) and services in (industrial)processes – including the consequences for a new creation of value, businessmodels as well as downstream services and work organisation. CPS refers tothe network connections between humans, machines, products, objects, andICT (information and communication technology) systems.” (Wisskirchen, etal., 2017: 12)

Overall, the collective power of select technologies to transform the automationlandscape amplifies multi-fold as compared to their power if taken individually. Thisamplified effect that we will call ‘The Great Convergence’ (GC) is depicted in figure1. The detailed building and placement of each component of the framework, as wellas individual evidence substantiating the maturity level of the component, has beendetailed in the next section.

The GC framework has been conceptualised as a building known as ‘IntelligentProcess Automation.’ It has got two layers viz. Foundation layer and the building orthe Enablement layer. The Foundation layer forms a conductive base for leveragingthe technologies of the Enablement layer. The Enablement layer, in turn, has beenmodeled around the key characteristics of AI technologies viz. learn, think, interact,and decide (TCS, 2017). Even though some of the earlier technologies included in theframework may not strictly conform to these characteristics, they do lead up to newertechnologies that imbibe these characteristics, thus providing a time-continuity in theframework.

The framework illustrates the continuum and convergence of technological progressindividually and collectively. It is important for us to understand that technologicaldisruption and changes happen incrementally on a continuous basis all around us.

The Foundation Layer

GC starts with the foundation layer comprising of:1. Data Management2. Hardware Infrastructure3. Communication Highway

Data management - According to a 2016 IBM estimate, around 90% of data in theworld today has been created in the last two years. A mind-boggling 2.5 quintillion (aquintillion in the US is a billion, i.e., 1018) bytes of data gets generated every day(Seagroatt, 2017). Data forms the input for any machine-led automation. Traditionally,routine manual tasks are automated by breaking them down into well-defined flowcharts and incorporating the same in software programs and scripts (Acemoglu &Autor, 2011) . Input triggers a pre-programmed action or a set of actions resulting inthe desired outcome.


Data management technology covering data storage, retrieval, and analysis usessystems with machine learning algorithms that can discover patterns in largeheterogeneous data sets, commonly called big data. Machine learning algorithms usethis big data as trainer data to mimic human understanding of context and experience(Frey & Osborne, 2013).

AIMA Journal of Management & Research, February 2020, Volume 14 Issue 1/4, ISSN 0974 – 497Copy right© 2020 AJMR-AIMA

Figure 1 : The Great Convergence - Towards Intelligent Process Automation


For example, taking the case of handwriting recognition, OCR technologies have beenaround for a long time but with limited practical uses. The true success of analgorithm for handwriting recognition is measured when it can handle data setscontaining different styles of writing (Plötz & Fink, 2009). This data, in turn, enrichesthe algorithm by providing the variety and the contingency elements a technologymust manage to form a reliable substitute for human work (Frey & Osborne, 2013),(Plötz & Fink, 2009). Intelligent OCR is now being used in processing servicerequests and account opening by banks (Manyika, et al., 2013).

Fraud detection is a typical financial services task that benefits from the pattern andtrend analysis of huge tranches of data (Chui, Manyika, & Miremadi, 2016). In healthcare, diagnostics tasks are being automated using trainer data. One recent example isMax Healthcare, which is training an AI tool to generate X-ray diagnostic reports, byfeeding it with millions of patient records and the know-how of radiologists as trainerdata (Singh, 2018).

Thus, with the progress in data technologies, automation is no longer confined toroutine tasks that can be written as rule-based software queries but is extending tonon-routine yet predictable tasks which can be translated into well-defined problemsusing algorithms working on now commonly available big data (Brynjolfsson &McAfee, 2011), (Frey & Osborne, 2013). This finding was also substantiated by ourexpert respondents.

Hardware infrastructure - The fuel of every technological engine is ‘computationalpower,’ and the start of every discussion on computational power is quite obviouslyMoore’s law1, which has incidentally held ground for decades. A 2011 study in thejournal Science showed that in recent times, every new year allowed humans to carryout roughly 60% more computation than possibly could have been executed by allexisting general-purpose computers in the year before. (Hilbert & López, 2011)

The exponential increase in computational power has been mirrored by a proportionaldecline in the unit cost of that power, as has been depicted in figure 2. Cheapercomputer power and storage that facilitates faster processing means a system that mayhave taken a few days to solve complex problems in the 1990s now takes a fraction ofthat time at a fraction of that cost.

2 Moore's law, prediction made by American engineer Gordon Moore in 1965 that thenumber of transistors per silicon chip doubles every year.(Source:https://www.britannica.com/technology/Moores-law accessed on February 12th, 2018)

https://en.wikipedia.org/wiki/Science_(journal)

https://www.britannica.com/technology/Moores-law


Figure 2: Computer Power

(Source: Nordhaus (2007): updated data through 2010 from Nordhaus, personalwebsite, http://www.econ.yale.edu/-Nordhaus/homepage/, “Two Centuries of

Productivity Growth in Computing.”)

The third and equally important aspect is the accessibility of high-end servers andhardware infrastructure. Easy access to virtually unlimited storage and processingpower is enabled via server farms and cloud technologies abundantly provided by thelikes of Amazon, Microsoft, Google, etc. (Gutierrez, Boukrami, & Lumsden, 2015) .Cloud models provide the means and availability to deploy power-hungry advancedmachine learning algorithms on large amounts of data to get quick outcomes.

Communication highway - The average internet connection speed in the USA hasgrown from 3.6 Mbps in 2007 to 18.7 Mbps in 2017, a staggering growth rate of over100% 2. Closer home, though ranked 89th globally, India was also recordingimpressive growth in average internet access speed from less than 1 Mbps in 2007 to6.5 Mbps in 2017 3. High-speed communication links and wireless technologiesprovide the highways for a seamless flow of information and data.

According to the study published in journal Science, the world’s capacity for 2-waytelecommunication evaluated between 1986 to 2007 grew at 28% per year. It has onlyrisen at a faster pace in the last decade since 2007 (Hilbert & López, 2011). Fastercommunication enables connected devices to transmit large amounts of information inreal-time to back-end systems for processing and for interaction systems tocommunicate back to front-end systems.

3 Source: https://www.statista.com/statistics/616210/average-internet-connection-speed-in-the-us/ accessed on February 17th, 2018

4 Source: https://dazeinfo.com/2017/06/02/india-internet-speed-growth-q1-2017/accessed on February 17th, 2018


While this is true in general, our panel of experts interviewed also mentionedcommunication technology or connectivity as a serious challenge in the IndianBanking context, especially in the non-urban centers where banking services arerequired to be delivered, and automation could help in doing that in a cost-effectiveand efficient manner. Connectivity remains a bottleneck for intelligent processautomation in banks in India.

1.1. The Enablement Layer

Leveraging the developments in the foundation layer is a set of technologies, whichcomprise the ‘enablement layer.’

Digital technologies (learning) - Earlier we have discussed, machine learningalgorithms being able to extract hidden value from millions of heterogeneous datapoints and sources to apply advanced predictive analytics techniques and unearthuseful patterns hereto impossible by humans (Brynjolfsson & McAfee, 2011).

Researchers from MIT Sloan School of Management have hypothesized that everytask eventually translates into judgment based on predictions, and hence moreaccurate predictions make for sound judgments (Agrawal, Gans, & Goldfarb, 2017) .If machines can generate reliable predictions and use those predictions to prescribethe next course of action, it will enable automation of tasks with a higher cognitiveelement. (Agrawal, Gans, & Goldfarb, 2017). Analytics technologies have advancedrapidly from the days of post-facto static reporting to dynamic real-time reporting andnow predictive and prescriptive analytics. All this, of course, powered by the greatfirepower of the foundation layer, as discussed earlier.

In Banking, for example, predictive analytics is extensively used in the areas of riskmanagement and fraud prevention. Mastercard deploys predictive AI to improve theoverall accuracy of real-time approvals and false declines of transactions. Thisreduces operational costs and increases customer shopping satisfaction levels(Stancombe et al., 2017). Probability of loss is another area where, by subjecting loanportfolios to trend and pattern analysis, advanced algorithms predict the probability ofdefault and assist in managing the credit risk of these portfolios (Seagroatt, 2017) .The interview responses confirmed the use of AI-powered algorithms in areas liketransaction monitoring, AML processing, alternate credit scoring models. Thisoutcome can also be seen abundantly in the use-case detailed in section 6 of thisarticle.

Readers & sensors (interaction) - OCR, bar code readers, and biometric recognitiontechnologies have been around for decades, albeit with limited discrete use in accesscontrol, authentication, document management, invoicing, etc. Current sensortechnologies falling under the umbrella of connected IoT devices augmented reality,image, speech, and biometric recognition has made sensor data one of the mostimportant sources of big data (Manyika, et al., 2013) . The net effect is a gradualmovement from discrete task-based use to integrated end-to-end deployments.

As per the Robotics Federation, the advanced sensors and readers thus form onecritical connection between reality (humans, documents, devices) and intelligentalgorithms/bots/systems running in the background. (Wisskirchen, et al., 2017) . This


combined effect of machine learning with an OCR engine can also be seen in the use-case we detail subsequently.

Natural language processing (interaction) – 19th century heralded the early ‘human-computer interaction’ technologies like Dictaphones and personal digital assistants(PDAs). The last few years have seen significant developments in the comprehension,processing, and generation of natural languages, the way we commonly see in thelikes of Alexa, Siri, Google NOW, and other such virtual assistants. (Chui, Manyika,& Miremadi, 2016) The key difference is the approach where earlier humans wereadapting to machine language, and now machines are adapting to human language,the later fostering faster and more user-friendly, 2-way interaction between man &machine. These virtual assistants are powered by technologies that can understandvarying accents, interpret conversational contexts, and respond intelligently (Manyika,et al., 2017a).

Such advances in user interfaces enable machines to manage and address a widerrange of human requests, thus allowing more non-routine cognitive jobs to becomeautomatable. For example, a company called Smart Action now provides callcomputerisation solutions that use machine language technology and advanced speechrecognition to improve upon conventional interactive voice response systems. (Frey &Osborne, 2013). Many services companies in India are also deploying chatbots andvoice assistants for customer services. The use of NLP and NLG is also seen in otherforms of communication like email, chats, etc. (FICCI; NASSCOM; Ernst & Young,2017).

However, there is one India specific challenge brought out by the experts, which isoften ignored in global studies. This is the multiplicity of Indian languages, dialects,and accents, which poses a problem in oral communication using NLP/NLG today.Globally, NLP/NLG deployments typically assume single language and a literatelanguage set of users, which is certainly not the case with a service like banking. Thisis another unique challenge to intelligent process automation in Indian Banking.

Artificial intelligence (thinking & decision-making) - The confluence of all layersdiscussed earlier; artificial intelligence comprises a basket of technologies, models,and algorithms. Four popular terminologies among these are machine learning, deeplearning reinforcement learning, and cognitive systems.

Machine learning coined by Arthur Samuel in 1959 is quite simply the fieldof computer science that gives computer systems the ability to “learn” (i.e.,progressively improve performance on a specific task) with data, without beingexplicitly programmed (Mitchell, 2006) . To be more precise, we say that a machinelearns with respect to a task, performance metric, and type of experience andimproves its performance at the same task following the same experience (Mitchell,2006).

This basket of technologies and algorithms, powered by the foundation layer andother enabling technologies, are the brainpower behind AI and are used in variedcontexts to simulate human tasks. The technologies on their own are fairly advancedbut face some constraints to their ubiquitous use, such as availability of adequately

https://en.wikipedia.org/wiki/Computer_science

https://en.wikipedia.org/wiki/Computer_systems

https://en.wikipedia.org/wiki/Data


tagged trainer data, sufficient experience of learning, doing, success & failure, and anarrow range of operation.

DISCUSSION - INTELLIGENT PROCESS AUTOMATION IN BANKING

Service automation has traditionally been an intensive human job requiring typicalhuman cognitive traits of perception, dexterity, reasoning, judgment, empathy, anddecision-making. (Agrawal, Gans, & Goldfarb, 2017) . Previous attempts at serviceautomation have essentially been as aids to humans in service delivery. It is typicallycharacterised by input in the form of a data point or information, which is then actedupon by a pre-programmed set of instructions to provide the output (Acemoglu &Autor, 2011).

Hence the traditional approach to task automation using computer-controlledequipment and following certain pre-defined logic and rules gets augmented with theadditional steps of learning, thinking, interacting and deciding as depicted in Figure 3.

Figure 3 : Process Automation Cycle in an AI-Driven Environment

An AI-driven system not only learns from each process/task iteration, but also fromthe contextual information of the process itself, much like human learning process.This learning empowers the system to think, decide and initiate at its discretion anyinteraction between systems and between man-machine for completion of the process.

In order to get an insight into how intelligent automation is unfolding in bankingservices, we asked our panel of experts the following questions:

Question (1): “how automation in the past differs from intelligent automationnow?”

Question (2): “Looking into the future, do you see more automation in the systemsthe bank uses for delivering its banking services over the next ten years? If yes, whichfunctions? If no, kindly explain in support of your answer”.


25 out of 28 experts cited the collective presence of data, computing power, andtechnology solutions as the fundamental difference in the ability to harness intelligenttechnologies, especially with reference to banking. The experts were asked to citeexamples to support their view. Customer servicing (n=18) and retail loan processing(n=23) were the most frequently cited functions, which, even though they are alreadyautomated, are being further automated using intelligent technologies.

We select the ‘Processing of a Retail Loan’ as a use case to further discuss the GCframework on Intelligent Process Automation. For the sake of simplicity, we take apersonal loan application being processed in a fairly automated environment andcompare that with the process being executed in the intelligent environment usingcomponents of GC. Exception tasks are excluded from the scope of the study as theyare handled by the human being in all scenarios. Figure 4 below is a level 0 processchart of a typical retail loan application approval process.

The key process steps depicted above highlight the following key features:

1. Automation is discretised using a loan processing system that does standardisedstandalone functions like calculate a credit score, store and forward loan file,workflow management, structured data uploads, and downloads.

2. The process is mainly orchestrated by a human being, the Loan ProcessingExecutive (LPE). The orchestration is done manually, whether it is logging intomultiple systems for credit checks, document verification, or interacting with arelationship manager, etc.

3. While the process is largely rule-driven, any special conditions which could beapplicable are manually understood by either the LPE or the approver andmanually incorporated in the sanction details.

4. Even though data may be available digitally, there is a large amount of duplicatedata entry in various interfacing systems, especially the 3rd party systems.Automation to the extent of manually downloading data from one system anduploading the same in another system may be available in some banks.

5. Documentation remains a paper-based, manual exercise.6. Credit scoring models rely on traditional parameters manually entered in the

system.

The same retail loan process in an intelligently automated environment will look asdepicted in figure 5 below:


(source: Primary data viz.expert interviews)


Figure 5 : Intelligent Automation process of a Retail Personal Loan Approval

(source: Primary data viz.expert interviews, https://www.unlockinsights.com/blog/use-cases-of-rpa-in-banking-industry/,https://thelabconsulting.com/robotics-in-banking-with-4-rpa-use-case-examples/)

https://www.unlockinsights.com/blog/use-cases-of-rpa-in-banking-industry/

https://thelabconsulting.com/robotics-in-banking-with-4-rpa-use-case-examples/


The key process steps in the intelligent automation context have the followingcharacteristics:

1. It is integrated process automation. For example, the handoffs between step2 to step 3 or from step 10 to 11 are automatically handled by the system,and data flows seamlessly from one system to another in an integratedmanner.

2. The process is mainly orchestrated by a BOT (a Conversational Interface,the equivalent of a robot in service setup) managing most part of theprocess, including interactions with multiple external entities. This reducesturnaround times, improves accuracy, and reduces the manual drudgery ofthe LPE, who was the orchestrator earlier.

3. BOT itself leverages multiple technologies to execute different tasks. Thisis where the convergence can be seen. For example, step 4 is executedusing OCR with machine learning (Reader plus AI), step 5 is againexecuted with a base set of rules enriched by machine learning from trainerdata (AI plus digital technologies), step number 11 uses natural languageprocessing & generation (Conversational Interface) to ‘write’ an email tothe customer/relationship manager.

4. Even the credit scoring in step 8 moves from traditional scoring models tomodern behavioural models (Digital Technologies on Big Data) where datais automatically extracted from various other external databases and socialmedia sites to create a more holistic profile of the borrower.

5. Trust-based digital document exchange is enabled by other intelligenttechnologies like Blockchain, thus making documentation an automatedtask.

6. Improved credit scoring and recommendations in step 9 are possible whenadvanced analytics work on huge volumes of structured and unstructureddata of trainer cases. These are fitted by machine learning algorithms tomake the recommendations.

7. The BOT exhibits human traits of reading, initiating, writing, sensing,deciding, and recommending; tasks earlier done by the LPE.

Thus, using the convergence of technology components, we can see how the processhas evolved from a simple Input-Output-Action to an intelligent process imbibinghuman-like traits. Our interviews with experts in banking and technology confirm thatthis scenario is not a futuristic scenario, and many such use-cases are playing out inbanks in India and globally even today.

CONCLUSION &WAY FORWARD

Oxford University professors Carl Frey & Michael Osborne, in their seminal paper onjob computerisation, have highlighted that one of the key factors driving automationwill be the technological advances that allow problems to be adequately specified andthese technological advances are also the boundaries for the extent of such automation(Frey & Osborne, 2013). Two significant factors namely 1) movement fromtechnology-literate people to people-literate technology and, 2) convergence of thepeaking of multiple technological factors are contributing to expanding technologicalboundaries (Manyika, et al., 2017b)


At the beginning of this article, we posed the quintessential question plaguingmankind, whether machines will cause mass human unemployment and set out beforeus the task of confirming that indeed the potential disruption is for real. We haveshown with the GC Framework for Intelligent Process Automation that all theecosystem elements critical to Artificial intelligence and related technologies haveemerged from ‘hype mode’ into ‘real business.’ Using the GC Framework, weconfirm the simultaneous maturity of the technology components, which has beenvalidated using primary qualitative data and demonstrated using a retail loan approvalprocess as a use-case. It further demonstrates that automation is now no longerconfined to computerisation of tasks following rule-based procedures but is spreadingto any task where data availability becomes a substitute for human knowledge andmachine & deep learning algorithms evolve to the ability to simulate human decisions.

Banks across the globe are implementing AI & Machine Learning in bankingoperations like account opening, loan processing, risk management, customer support,fraud detection, NPA management, etc., and other tasks that are highly process-oriented and data-intensive (Urs, 2017).

One of the surprise findings from discussion with experts was that bankers (both inthe public sector and private sector) though having the positional power to maketechnology decisions do not have enough knowledge and understanding of thesetechnologies. This is a hurdle faced by senior bankers making them averse to makingany significant decisions regarding adopting intelligent automation technologies intheir core business operations. This article seeks to dispel that confusion amongbankers by laying out the various components in a logical, easy to comprehendframework. It is through this article the authors want to initiate discussions in thebanking industry with the objective of reviewing processes and proactively addressingthe upcoming changes to the benefit of the stakeholders.

LIMITATIONS

Any study in the futuristic area of technology and process automation is a qualifiedexpectation of experts and decision-makers in the field. It is neither a time-seriesstudy nor a statistical study of the before/after scenario simply because the events areunfolding as we speak. Thus, it needs to be viewed with pragmatic latitude andskepticism alike.

Also, the GC framework is not strictly a comprehensive time-series graph or a pointin time depiction of the maturity of these technologies. There are many moreconstituent technologies and factors contributing to automation, and there could beoverlaps in their development times and maturity levels.

For simplicity, the framework captures only critical milestones in process automation.There are many more emerging and/or adjacent technologies such as Blockchainwhich have not been captured in this framework. Some of these technologies couldhave an equally disruptive effect on automation but are currently out of the scope ofour discussion here. However, this exclusion does not dilute the main argumentfavouring the convergence effect.


REFERENCESAcemoglu, D., & Autor, D. (2011). Skills, Tasks and Technologies: Implications for

Employment and Earnings. Handbook of Labor Economics, Volume 4, Part B,Pages 1043-1171. doi:https://doi.org/10.1016/S0169-7218(11)02410-5

Acemoglu, D., & Robinson, J. A. (2012). The Turning Point. In Why nations fail: theorigins of power, prosperity, and poverty (p. 109). Random House Digital, Inc.Retrieved January 13, 2018, fromhttps://archive.org/stream/WhyNationsFailDaronAcemogluAndJamesA.Robinson1/Why+Nations+Fail+-+Daron+Acemoglu+and+James+A.+Robinson+%281%29_djvu.txt

Agrawal, A., Gans, J. S., & Goldfarb, A. (2017, Spring). What to Expect fromArtificial Intelligence. MIT Sloan Management Review, Vol. 58, No. 3, pp. 22-26. Retrieved January 23, 2018, from http://mitsmr.com/2jZdf1Y

Botha, N., & Atkins, K. (2005). An assessment of five different theoreticalframeworks to study the uptake of innovations. Paper presented at the 2005NZARES Conference. NZARES. Retrieved May 25 2018, fromhttps://pdfs.semanticscholar.org/c85a/325bc004cb4ba9c2fa7f59ae7139aec92397.pdf

Brynjolfsson, E., & McAfee, A. (2011). Race against the machine: How the digitalrevolution is accelerating innovation, driving productivity, and irreversiblytransforming employment and the economy. Digital Frontier Press. RetrievedDecember 24, 2017, fromhttp://ebusiness.mit.edu/research/Briefs/Brynjolfsson_McAfee_Race_Against_the_Machine.pdf

Chui, M., Manyika, J., & Miremadi, M. (2016). Where machines could replacehumans—and where they can’t (yet). McKinsey Quarterly, Online. RetrievedDecember 26, 2018, from https://www.mckinsey.com/business-functions/digital-mckinsey/our-insights/where-machines-could-replace-humans-and-where-they-cant-yet

Dawar, S., & Lacy, P. (2017). Leading in the New: Enabling Advanced Technology-based Innovation. Retrieved from Accenture: https://www.accenture.com/in-en/insight-innovate-in-the-new?sr_source=lift_twitter&c=dig_abjify17_10000004&n=smc_0717#6

Economist. (2015, May). The Dawn of Artificial Intelligence. The Economist.Retrieved January 13, 2018, fromhttps://www.economist.com/news/leaders/21650543-powerful-computers-will-reshape-humanitys-future-how-ensure-promise-outweighs

FICCI; NASSCOM; Ernst & Young. (2017). Future of jobs in India : A 2022perspective. FICCI. Retrieved December 26, 2017, fromhttp://www.ey.com/Publication/vwLUAssets/ey-future-of-jobs-in-india/FILE/ey-future-of-jobs-in-india.pdf

Frey, C. B., & Osborne, M. A. (2013, September 17). The Future of Employment:How Susceptible Are Jobs to Computerisation? Technological Forecastingand Social Change, 254-280.doi:https://doi.org/10.1016/j.techfore.2016.08.019.

Gangwar, H., & Date, H. (2015, February). Understanding determinants of cloudcomputing adoption using an integrated TAM-TOE model. Journal ofEnterprise Information Management, 28(1). doi:DOI: 10.1108/JEIM-08-2013-0065


Gartner Inc. (2018, July 24). Hype Cycle for Artificial Intelligence, 2018. RetrievedJuly 28, 2018, from Gartner.com: https://www.gartner.com/home

Gartner Inc. (2017, July 28). Hype Cycle for Analytics and Business. Retrieved May25, 2018, from Gartner.com: https://www.gartner.com/home

Gartner Inc. (2017, July 21). Hype Cycle for Emerging Technologies, 2017. RetrievedMay 25, 2018, from gartner.com: https://gartner.com/home

Gutierrez, A., Boukrami, E., & Lumsden, R. (2015). Technological, organisationaland environmental factors influencing influencing managers’ decision to adoptcloud computing in the UK. Journal of Enterprise Information Management,Vol. 28 No. 6, pp. 788-807. doi:10.1108/JEIM-01-2015-0001

Hafner, J. (2018, February 27). California approves fully driverless cars on publicroads for testing. Retrieved February 28, 2018, from www.usatoday.com:https://www.usatoday.com/story/money/nation-now/2018/02/27/california-approves-fully-driverless-cars-public-roads-testing/378703002/

Hilbert, M., & López, P. (2011, April). The world’s technological capacity to store,communicate, and compute information. Science. doi:doi:10.1126/science.1200970.

HOTI, E. (2015). The technological, organizational and environmental framework ofis innovation adaption in small and medium enterprises evidence fromresearch over the last 10 years. International Journal of Business andManagement, III(4/2015), 1-14. doi:10.20472/BM.2015.3.4.001

Ismail, N., & Linas, P. (2017, June 6). Busting three AI myths to help make businesssmarter: AI is still far in the future, AI isn’t going to have a big impact onwork and AI will replace human employees. Retrieved August 25, 2017, fromInformation Age: http://www.information-age.com/making-business-smarter-3-misconceptions-ai-123466632/

Jayanthi, C. V., & Pooja, S. (2011). Indian Software Industry - Moving up the ValueChain and Managing Non-Linear Growth. CSITM Participatory Workshop.Bangalore: Indian Institute of Management. Retrieved February 13, 2018,fromhttp://www.iimb.ac.in/sites/default/files/u181/Indian%20Software%20Industry%20-%20Managing%20non-linear%20growth%20-%20id%201055-1057_0.pdf

Kaftzan, J. (2017, August 7). Artificial intelligence is transforming the enterprise.Retrieved from Information Age: http://www.information-age.com/artificial-intelligence-transforming-enterprise-123467724/

LiveMint. (2017, August 17). Demand for artificial intelligence, machine learningexperts to rise 60% by 2018: report. Retrieved August 23, 2017, fromLiveMint:http://www.livemint.com/Technology/rYudcXx0AA9UPXABa9ooFL/Demand-for-artificial-intelligence-machine-learning-experts.html

Manyika, J., Chui, M., Bughin, J., Dobbs, R., Bisson, P., & Marrs, A. (2013).Disruptive technologies: Advances that will transform life, business, and theglobal economy. McKinsey Global Institute. Retrieved August 4, 2018, fromhttps://www.mckinsey.com/~/media/mckinsey/business%20functions/mckinsey%20digital/our%20insights/disruptive%20technologies/mgi_disruptive_technologies_full_report_may2013.ashx

Manyika, J., Chui, M., Miremadi, M., Bughin, J., George, K., Willmott, P., &Dewhurst, M. (2017a). A Future That Works: Automation, Employment, AndProductivity. McKinsey Global Insitute. Retrieved January 31, 2018, from


https://www.mckinsey.com/~/media/McKinsey/FeaturedInsights/DigitalDisruption/Harnessingautomationforafuturethatworks/MGI-A-future-that-works_Full-report.ashx

Manyika, J., Lund, S., Chui, M., Bughin, J., Woetzel, J., Batra, P., & Ko, R. (2017b).Jobs Lost, Jobs Gained: Workforce Transitions in a Time of Automation.McKinsey Global Institute. Retrieved May 5, 2018, fromhttps://www.mckinsey.com/~/media/mckinsey/featuredinsights/FutureofOrganizations/Whatthefutureofworkwillmeanforjobsskillsandwages/MGI-Jobs-Lost-Jobs-Gained-Report-December-6-2017.ashx

McKinsey. (2017). A Future That Works: The Impact Of Automation In Denmark.The Tuborg Research Centre For Globalisation & McKinsey & Company.Retrieved December 24, 2017, from https://www.mckinsey.com/denmark/our-insights/a-future-that-works-the-impact-of-automation-in-denmark

Mitchell, T. M. (2006, July). Discipline of Machine Learning. doi:CMU-ML-06-108Nilsson, N. J. (1984, Summer). Artificial Intelligence, Employment and Income. The

AI Magazine, 5-14. Retrieved January 12, 2018, fromhttp://ai.stanford.edu/~nilsson/OnlinePubs-Nils/General%20Essays/AIMag05-02-002.pdf

Olechowski, A., Eppinger, S. D., & Joglekar, N. (2015). Technology ReadinessLevels at 40: A Study of State-of-the-Art Use, Challenges, and Opportunities.PICMET ’15: Management of the Technology Age, (pp. 2084-2094). RetrievedMay 25, 2018, fromhttp://web.mit.edu/eppinger/www/pdf/Eppinger_PICMET2015.pdf

Plötz, T., & Fink, G. A. (2009). Markov models for offline handwriting recognition:A Survey. International Journal on Document Analysis and Recognition,12(4), 269 -298. doi:https://doi.org/10.1007/s10032-009-0098-4

Seagroatt, M. (2017). What are the Applications for Artificial Intelligence inSecurities Finance and Collateral Management. Broadridge. RetrievedDecember 27, 2017, from https://www.broadridge.com/white-paper/what-are-the-applications-for-artificial-intelligence-in-securities-finance-and-collateral-management

Singh, S. (2018, January 23). Will Artificial Intelligence take over jobs? . RetrievedJanuary 24, 2018, from www.economictimes.indiatimes.com://economictimes.indiatimes.com/articleshow/62610145.cms?utm_source=contentofinterest&utm_medium=text&utm_campaign=cppst

Stancombe, C., Tolido, R., Thieullent, A.-L., Buvat, J., Subrahmanyam, K., Khadikar,A., & Chandna, A. (2017). Turning AI into concrete value. CapGemini.Retrieved September 5, 2017, fromhttps://www.capgemini.com/resources/artificial-intelligence-where-and-how-to-invest/

TCS. (2017). Getting Smarter by the Sector: How 13 Global Industries Use ArtificialIntelligence. Tata Consultancy Services. Retrieved January 12, 2018, fromhttps://www.tcs.com/global-trend-study-artificial-intelligence-reveals-industry-wide-investment-2020

Understanding the Evolution and Importance of Business Process Automation. (2018,January 15). Retrieved from www.smartsheet.com:https://www.smartsheet.com/understanding-evolution-and-importance-business-process-automation

Urs, A. (2017, August 13). Banks exploring use of artificial intelligence in operations.Retrieved August 27, 2017, from Hindu Business Online:


http://www.thehindubusinessline.com/money-and-banking/banks-exploring-use-of-artificial-intelligence-in-operations/article9816427.ece

Wisskirchen, G., Blandine , B. T., Bormann, U., Muntz, A., Niehaus, G., Jiménez , G.,& Brauchitsch, , B. v. (2017). Artificial Intelligence and Robotics and TheirImpact on the Workplace. IBA Global Employment Institute. RetrievedDecember 27, 2017, fromhttps://www.ibanet.org/Document/Default.aspx?DocumentUid=c06aa1a3-d355-4866-beda-9a3a8779ba6e


Annexure 1Gartner is a leading research and consulting company specialising in technologyrelated areas. Gartner Hype Cycles provide a graphic representation of the maturityand adoption of technologies and applications, and how they are potentially relevantto solving real business problems and exploiting new opportunities. In these reportsGartner also defines the maturity levels of various business technologies on a 7 stageclassification as detailed here:Maturity Level StatusEmbryonic In labsEmerging Commercialization by vendors

Pilots and deployments by industry leadersAdolescent Maturing technology capabilities and

process understanding Uptake beyond early adopters

Early mainstream Proven technology Vendors, technology and adoption rapidly

evolvingMaturemainstream

Robust technology Not much evolution in vendors or

technologyLegacy Not appropriate for new developments

Cost of migration constrains replacementObsolete Rarely used(source: Hype Cycle for Emerging Technologies, 2017, Gartner id: G00314560,published 21 July 2017)

Documents

A I -D A –I I R Tapps.aima.in/ejournal_new/articlesPDF/3-Pooja-Sharma.pdfMany other studies have also defined AI in a similar manner comprising technologies like - speech recognition,